In [1]:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
import seaborn as sns
from sklearn.preprocessing import Imputer
from sklearn.model_selection import train_test_split
from sklearn import metrics
from sklearn.tree import DecisionTreeClassifier
In [2]:
df=pd.read_csv("vehicle.csv")
In [3]:
df.head()
Out[3]:
In [4]:
df.shape
Out[4]:
In [5]:
df.info()
In [6]:
df.describe().transpose()
Out[6]:
In [7]:
df_copy=df.copy()
df_copy.shape
Out[7]:
In [8]:
# We can see that there are 'NaN' values in few columns.
df_copy.isna().sum()
Out[8]:
In [9]:
# attribute that contains Zero values.
print((df_copy == 0).sum())
In [10]:
"""
<b>Now skewness_about,skewness_about.1 contains 77 and 30 Zero(0) values respectively.Earlier though was to replace zeros but
i think these are not errors.so will keep them as it.<b>
"""
Out[10]:
In [11]:
print((df_copy==0).sum())
In [12]:
df_copy.info()
In [13]:
# Now we will check total NaN values in all columns
df_copy.isna().sum()
Out[13]:
In [14]:
"""
We can see we have columns with 'NaN' values.We will replace them with mean/median accordingly.
circularity 5
distance_circularity 4
radius_ratio 6
pr.axis_aspect_ratio 2
scatter_ratio 1
elongatedness 1
pr.axis_rectangularity 3
scaled_variance 3
scaled_variance.1 2
scaled_radius_of_gyration 2
scaled_radius_of_gyration.1 4
skewness_about 6
skewness_about.1 1
skewness_about.2 1
"""
"""
I can't plot distplot as columns contains NaN values.So i will remove NaN and
then check the data is normally distributed or not
"""
# df_copy_circularity=df_copy['circularity'].dropna(axis=0)
# df_copy_distance_circularity= df_copy['distance_circularity'].dropna(axis=0)
# df_copy_radius_ratio= df_copy['radius_ratio'].dropna(axis=0)
# df_copy_praxis_aspect_ratio= df_copy['pr.axis_aspect_ratio'].dropna(axis=0)
# df_copy_scatter_ratio= df_copy['scatter_ratio'].dropna(axis=0)
# df_copy_elongatedness= df_copy['elongatedness'].dropna(axis=0)
# df_copy_praxis_rectangularity= df_copy['pr.axis_rectangularity'].dropna(axis=0)
# df_copy_scaled_variance= df_copy['scaled_variance'].dropna(axis=0)
# df_copy_scaled_variance1= df_copy['scaled_variance.1'].dropna(axis=0)
# df_copy_scaled_radius_of_gyration= df_copy['scaled_radius_of_gyration'].dropna(axis=0)
# df_copy_scaled_radius_of_gyration1= df_copy['scaled_radius_of_gyration.1'].dropna(axis=0)
# df_copy_skewness_about= df_copy['skewness_about'].dropna(axis=0)
# df_copy_skewness_about1= df_copy['skewness_about.1'].dropna(axis=0)
# df_copy_skewness_about2= df_copy['skewness_about.2'].dropna(axis=0)
# sns.distplot(df_copy_circularity)
# skewness_about=df_copy['skewness_about'].dropna()
# skewness_about.count()
Out[14]:
In [15]:
# Class is categorical.will convert it to interger and then replace NaN.
df_copy['class']=df_copy['class'].astype('category').cat.codes
df_copy['class'].head()
Out[15]:
In [16]:
# Missing value changed by using Impurter or by using fillNa method.
# imputer=Imputer(missing_values='NaN',strategy='median',axis=0)
# transformed_X=imputer.fit_transform(df_copy)
# transformed_X.dtype
df_copy.isnull().sum()
Out[16]:
In [17]:
# Missing value changed by using Impurter or by using fillNa method.
# df_copy_With_No_NA=df_copy.fillna(df_copy.median,inplace=True)
# df_copy_With_No_NA
df_copy_With_No_NAN_Values=df_copy.fillna(df_copy.median())
df_copy_With_No_NAN_Values.isnull().sum()
Out[17]:
Missing values are handeled by replacing with Median in the datasets.
Finding Outliers and Removing them¶
In [18]:
df_copy_With_No_NAN_Values.boxplot(figsize=(100,10))
Out[18]:
In [19]:
"""
We have outliers in radius_ratio,pr.axis_aspect_ratio,
max.length_aspect_ratio,scaled_variance,scaled_radius_of_gyration.1,skewness_about,skewness_about1
"""
Out[19]:
In [20]:
Q1 = df_copy_With_No_NAN_Values.quantile(0.25)
Q3 = df_copy_With_No_NAN_Values.quantile(0.75)
IQR = Q3 - Q1
print(IQR)
In [21]:
df_copy_With_No_NAN_Values = df_copy_With_No_NAN_Values[~((df_copy_With_No_NAN_Values < (Q1 - 1.5 * IQR)) |(df_copy_With_No_NAN_Values > (Q3 + 1.5 * IQR))).any(axis=1)]
df_copy_With_No_NAN_Values.shape
Out[21]:
In [22]:
"""
We can see in boxplot,outliers are removed
"""
df_copy_With_No_NAN_Values.boxplot(figsize=(100,10))
Out[22]:
Now we will find correlation to find relation between Independent Var¶
In [23]:
df_copy_With_No_NAN_Values_Corr=df_copy_With_No_NAN_Values.corr(method='pearson')
df_copy_With_No_NAN_Values_Corr
Out[23]:
In [24]:
# plot the heatmap
sns.heatmap(df_copy_With_No_NAN_Values_Corr,
xticklabels=df_copy_With_No_NAN_Values_Corr.columns,
yticklabels=df_copy_With_No_NAN_Values_Corr.columns)
Out[24]:
In [25]:
cmap = cmap=sns.diverging_palette(5, 250, as_cmap=True)
def magnify():
return [dict(selector="th",
props=[("font-size", "7pt")]),
dict(selector="td",
props=[('padding', "0em 0em")]),
dict(selector="th:hover",
props=[("font-size", "12pt")]),
dict(selector="tr:hover td:hover",
props=[('max-width', '200px'),
('font-size', '12pt')])
]
df_copy_With_No_NAN_Values_Corr.style.background_gradient(cmap, axis=1)\
.set_properties(**{'max-width': '80px', 'font-size': '10pt'})\
.set_caption("Hover to magify")\
.set_precision(2)\
.set_table_styles(magnify())
Out[25]:
In [26]:
# Find inferences
"""
Compactness shows high significance with circularity,distance_circularity,radius_ratio,
scatter_ratio,elongatedness in a negative way,pr.axis_rectangularity,max.length_rectangularity,
scaled_variance,scaled_variance.1
"""
#Let us check for pair plots
sns.pairplot(df_copy_With_No_NAN_Values,diag_kind='kde')
Out[26]:
3. Use PCA from scikit learn and elbow plot to find out reduced number of dimension (which covers more than 95% of the variance) - 10 points¶
Create a covariance matrix for identifying Principal components¶
In [26]:
# In this class is a dependent variable.So removing it for PCA.
X=df_copy_With_No_NAN_Values.drop(['class'],axis=1)
X.head()
y=df_copy_With_No_NAN_Values[['class']]
sns.pairplot(X,diag_kind='kde')
Out[26]:
In [27]:
"""
We will scale the data with the help of zscore.It is used to remove the influence of one attribute over others in terms of
units.
"""
from scipy.stats import zscore
XScaled=X.apply(zscore)
XScaled.head()
Out[27]:
In [28]:
# Create covariance matrix
cov_matrix = np.cov(XScaled,rowvar=False)
print('Covariance Matrix \n%s', cov_matrix)
Identify eigen values and eigen vector¶
In [29]:
# Step 2- Get eigen values and eigen vector
eig_vals, eig_vecs = np.linalg.eig(cov_matrix)
print('Eigen Vectors \n%s', eig_vecs)
print('\n Eigen Values \n%s', eig_vals)
Find variance and cumulative variance by each eigen vector¶
In [30]:
tot = sum(eig_vals)
var_exp = [( i /tot ) * 100 for i in sorted(eig_vals, reverse=True)]
cum_var_exp = np.cumsum(var_exp)
print("Cumulative Variance Explained", cum_var_exp)
In [31]:
plt.plot(var_exp)
Out[31]:
In [32]:
# Ploting
plt.figure(figsize=(10 , 5))
plt.bar(range(1, eig_vals.size + 1), var_exp, alpha = 0.5, align = 'center', label = 'Individual explained variance')
plt.step(range(1, eig_vals.size + 1), cum_var_exp, where='mid', label = 'Cumulative explained variance')
plt.ylabel('Explained Variance Ratio')
plt.xlabel('Principal Components')
plt.legend(loc = 'best')
plt.tight_layout()
plt.show()
In [33]:
"""
Visually we can observe that their is steep drop in variance explained with increase in number of PC's.
We will proceed with 7 components here.
"""
Out[33]:
Using PCA command from sklearn and find Principal Components. Transform data to components formed¶
In [34]:
# NOTE - we are generating only 7 PCA dimensions (dimensionality reduction from 18 to 7)
from sklearn.decomposition import PCA
pca7 = PCA(n_components=7)
pca7.fit(XScaled)
print(pca7.components_)
print(pca7.explained_variance_ratio_)
Xpca7=pca7.transform(XScaled)
In [40]:
"""
We can see these are independent and not corelated.So we have reduced dimensions
and covered almost 95 % of variation or information as rest are noise.
"""
sns.pairplot(pd.DataFrame(Xpca7))
Out[40]:
SVM¶
In [35]:
from sklearn import svm
In [ ]:
In [36]:
XScaled.shape
Out[36]:
In [37]:
y.head()
Out[37]:
In [38]:
# Split the Data.Considering all variables.
X_train,X_test,y_train,y_test=train_test_split(XScaled,y,test_size=.30, random_state=1)
In [39]:
# Gamma is a measure of influence of a data point.C is complexity of the model,lower C value creates simple hypersurface.
clf=svm.SVC(gamma=0.025,C=3)
clf
Out[39]:
In [40]:
clf.fit(X_train,y_train)
Out[40]:
In [41]:
y_predSVM=clf.predict(X_test)
In [42]:
from sklearn.metrics import accuracy_score
accuracy=accuracy_score(y_test,y_predSVM)*100
print(('Accuracy of SVM model with all variables is equal '+ str(round(accuracy, 2))) + ' %.')
In [43]:
# MAking confusion matrix
from sklearn.metrics import confusion_matrix
cmSVMAllVar=confusion_matrix(y_test,y_predSVM)
cmSVMAllVar
Out[43]:
SVM With PCA(7 Var)¶
In [44]:
Xpca7.shape
Out[44]:
In [45]:
# Split the data
X_train7Var,X_test7Var,y_train7Var,y_test7Var=train_test_split(Xpca7,y,test_size=.30,random_state=1)
In [46]:
clf7Var=svm.SVC(gamma=0.25,C=3)
clf7Var
Out[46]:
In [47]:
clf7Var.fit(X_train7Var,y_train7Var)
Out[47]:
In [48]:
y_PredSVM7Var=clf7Var.predict(X_test7Var)
y_PredSVM7Var
Out[48]:
In [49]:
from sklearn.metrics import accuracy_score
accuracy7Var=accuracy_score(y_test7Var,y_PredSVM7Var)*100
print(('Accuracy of SVM model with 7 variables is equal '+ str(round(accuracy7Var, 2))) + ' %.')
In [50]:
# MAking confusion matrix
from sklearn.metrics import confusion_matrix
cmSVM7Var=confusion_matrix(y_test7Var,y_PredSVM7Var)
cmSVM7Var
Out[50]:
Accuracy Comparison of SVM With or Without PCA¶
In [51]:
print(('Accuracy of SVM model with all variables is equal '+ str(round(accuracy, 2))) + ' %.')
print(('Accuracy of SVM model with 7 variables is equal '+ str(round(accuracy7Var, 2))) + ' %.')
Naive Bayes¶
In [52]:
from sklearn.naive_bayes import GaussianNB
classifier=GaussianNB()
In [57]:
# With all variables
classifier.fit(X_train,y_train)
y_predNaiveAllVar=classifier.predict(X_test)
y_predNaiveAllVar
cmNaiveAllVar=confusion_matrix(y_test,y_predNaiveAllVar)
cmNaiveAllVar
accuracyNaiveBayesAllVar=accuracy_score(y_test,y_predNaiveAllVar)*100
print('Accuracy of Naive Bayes for all Variables is equal ' + str(round(accuracy, 2)) + ' %.')
In [58]:
# With 7 variables
classifier.fit(X_train7Var,y_train7Var)
y_predNaive7Var=classifier.predict(X_test7Var)
y_predNaive7Var
cmNaive7Var=confusion_matrix(y_test7Var,y_predNaive7Var)
cmNaive7Var
accuracyNaiveBayes7Var=accuracy_score(y_test7Var,y_predNaive7Var)*100
print('Accuracy of Naive Bayes for 7 Variables is equal ' + str(round(accuracy, 2)) + ' %.')
Accuracy of Naive Bayes With or Without PCA¶
In [59]:
print('Accuracy of Naive Bayes for 7 Variables is equal ' + str(round(accuracy, 2)) + ' %.')
print('Accuracy of Naive Bayes for all Variables is equal ' + str(round(accuracy, 2)) + ' %.')
In [ ]:
In [60]:
# Hyper Parameter
In [61]:
from sklearn.metrics import classification_report
In [63]:
from sklearn.svm import SVC
In [64]:
print(classification_report(y_test, y_predSVM))
In [65]:
from sklearn.model_selection import GridSearchCV
# defining parameter range
param_grid = {'C': [0.1, 1, 10, 100, 1000],
'gamma': [1, 0.1, 0.01, 0.001, 0.0001],
'kernel': ['rbf']}
grid = GridSearchCV(SVC(), param_grid, refit = True, verbose = 3)
# fitting the model for grid search
grid.fit(X_train, y_train)
Out[65]:
In [66]:
# print best parameter after tuning
print(grid.best_params_)
# print how our model looks after hyper-parameter tuning
print(grid.best_estimator_)
In [67]:
grid_predictions = grid.predict(X_test)
# print classification report
print(classification_report(y_test, grid_predictions))
In [ ]: